The Interaction of Morphology and Syntax in Aﬃx Orderhome.uni-leipzig.de/jtrommer/papers/ym2.pdf · The Interaction of Morphology and Syntax in Aﬃx Order Jochen Trommer 1 Introduction

The Interaction of Morphology and Syntax inA!x Order

Jochen Trommer

1 Introduction

In this article, I propose a constraint-based account of ordering restrictions onsubject agreement a!xes. Di"erent orderings of subject agreement, crosslin-guistically and in single languages, are captured by di"erent rankings of uni-versal well-formedness constraints in the sense of Optimality Theory (OT,Prince and Smolensky 1993). Assuming a postsyntactic morphology moduleas in Distributed Morphology (DM, Halle and Marantz 1993), I argue thatpart of these constraints are purely morphological while others require thecorrespondence of a!x order to the underlying syntactic structure. Con-straint interaction of this type obviates any stipulation for whether singleagreement a!xes appear as prefixes or su!xes, and reduces seemingly id-iosyncratic facts to general principles. The theoretical results are based onempirical findings from a corpus of about 100 languages with subject agree-ment. The sample as well as the analysis focus on agreement patterns wheresubject agreement is split into person and number markers. Data of this typehave been largely ignored in the literature on a!x order, but prove essentialfor establishing morphological constraints whose e"ects can also be found in“standard” types of agreement.

1.1 A!xal Status and Alignment

What is systematic about a!x order in the following data from Georgian?1

(1) v-xedav v-xedav-t xedav-s xedav-enS1-see S1-see-PL see-S3s see-S3p‘I see’ ‘we see’ ‘he sees’ ‘they see’

1

Distributed Morphology, like most current accounts of inflectional morphol-ogy, would state for each a!x whether it is a prefix (v-) or a su!x (-t, -d,-en). This is unsatisfying even under a purely descriptive perspective. Amore parsimonious account would state that subject agreement a!xes inGeorgian are generally su!xes with the exception of v, which is a prefix.This interaction of a general rule and an exception statement can be natu-rally formalized by ranked violable constraints as in (2), where (2b) can onlyapply if (2a) does not

(2) a. v is a prefixb. Subject agreement a!xes are su!xes

These constraints are ranked because the higher ranked constraint (2a) hasto be obeyed in case of conflict, and violable since (2b) is then violated. Butwe can still do better and ask if there is a deeper reason why v- is a prefixbut not -t or -en. Indeed, looking at the languages of the world (cf. section4), one finds a general tendency for number agreement to be marked on theright and for person agreement to be marked on the left edge of the word.Thus (2) can be replaced by (3):

(3) a. Number agreement should be maximally rightwardsb. Person agreement should be maximally leftwards

Constraints of this type, which require minimal distance between edges (in(3a) between the right edge of number a!xes and the right edge of theword, in (3b) between the left edge of person a!xes and the left edge of theword), are called “alignment constraints” in the OT-literature (McCarthyand Prince 1993)2 and are independently motivated by applications in a bignumber of other linguistic domains (see section 4.4.2). The constraints in (3)clearly predict the order in v-xedav-t, because v- marks only person (1st) and-t only number (plural), but what about -en and -s? Since these mark personand number, both constraints are relevant. Again, since (3a) is ranked higherthan (3b) they are rightwards. Note that we have now completely reversedour original viewpoint: The position of v-, which at first seemed to be an id-iosyncratic fact about a single a!x, now appears as a phenomenon familiarfrom the literature on Optimality Theory, namely Emergence of the Un-marked (McCarthy and Prince 1994): Universal wellformedness-constraints(in this case (3b)) become visible in a language only under restricted circum-stances (the unique featural content of v- as pure person marker), blocked

2

otherwise by higher-ranked constraints (here (3a)).Following the logic of OT, constraints, while themselves universal, can

be ranked di"erently in di"erent languages. Thus we would also expect alanguage where pure number and pure person markers are positioned as inGeorgian but where mixed markers are prefixes. As we will see in section 4,this prediction is borne out in other languages such as Muna and Amharic.

However data from Island Kiwai (Wurm 1975) show that alignment can-not be the only crucial factor to account for the order of agreement a!xes:3

(4) a. Present b. Indefinite Futuren-V-duru-do ni-du-do-V-ri1-V-Tns-Du 1-Tns-Du-V-Tns

While the dual marker do is a su!x in the present tense (4a), as expected,it is a prefix in the indefinite future (4b). Note however that -do is alwaysright-adjacent to a Tense a!x (-duru in (4a), du- in (4b)). I propose toconnect this observation to the fact that in most languages agreement seemsto be adjoined to a Tense head (Halle and Marantz 1993: 146). As we willsee in section 6, the interaction of a further constraint, REFLECT(Agr) –requiring that the order of agreement a!xes reflects the position of theirunderlying syntactic host – with alignment constraints accounts for the factsin Kiwai as well as for further crosslinguistic generalizations on the order ofsubject agreement a!xes.

1.2 Overview

In section 2, I will sketch the formal framework I assume. Section 3 in-troduces the language sample I use. The results on person and numberagreement will be presented in section 4, where I also develop an analysisin terms of alignment constraints. In section 5, I will show, based on Julien(2000), that syntactic heads such as Tense and Aspect follow quite di"erentprinciples which result from the nature of possible movement operations insyntax. In the following two sections, it is shown how syntactic and morpho-logical constraints interact to derive the order of Tense and agreement in thegrammars of single languages (section 6) and in form of universal tendenciesin a!x order (section 7). In section 8, I argue that the analysis is consis-tent with findings on the relation of a!x order and basic word order, andsection 9 shows that it is empirically superior to alternative approaches to

3

a!x order. The paper concludes with some speculations on the asymmetrybetween morphology and syntax in a!x order (section 10). An appendix liststhe languages of my language sample including genetic and areal a!liationand the sources used.

2 The Framework

The framework I assume in this paper is Distributed Optimality (DO, Trom-mer 2002a, Trommer 2002b). DO is a constraint-based, modular version ofDistributed Morphology (DM, Halle and Marantz 1993). The name Dis-tributed Optimality refers to the assumption of di"erent morphosyntacticcomponents, all mapping specific inputs to outputs according to the prin-ciples of Optimality Theory (OT, Prince and Smolensky 1993, McCarthyand Prince 1994). In section 2.1, I briefly discuss the architecture of thegrammar in DO. In 2.2, I introduce some basic notions of OT and alignmentconstraints as the basic constraint type to account for the order of agreementa!xes.

2.1 The Architecture of the Grammar

In DO, morphosyntax involves the following three modules:

(5) a. Syntax (lexical items ! syntactic chains)b. Chain Interpretation (syntactic chains ! single heads)c. Head Interpretation (heads ! vocabulary items)

The Syntax component creates abstract syntactic representations which con-tain neither agreement nor case a!xes (while chains might be assigned ab-stract case). Crucially, Syntax creates chains from lexical items. Chain Inter-pretation maps chains onto single heads (put another way, traces, i.e., copiesproduced by movement are eliminated) and adds (abstract) case and agree-ment heads. As Syntax, this module does not involve phonological features.Finally, in Head Interpretation, the abstract heads from Chain Interpretationare mapped to (possibly underspecified) Vocabulary Items which introducephonological material. (6) illustrates this model for the German sentence wirtrinken, “we drink”:

(6) a. [+D+1+pl]i [+Tns] [[+D+1+pl]i [+V]]V P

4

b. [+D+1+pl][+Nom]] [+Tns] [[+D+1+pl] [[+V][+Agr+1+pl]]]V P

c. wir:[+D+1+pl] trink:[+V] en:[+Agr+pl]

The syntax generates the representation in (6a). There is a chain comprisingtwo instances of the pronoun ([+D+1+pl]), one in the specifier of IP and theother in its base position in the VP.4 Chain Interpretation maps this chainonto (6b), where the chain is reduced to the single head in Spec IP to whichthe [+Nom] head is added. To V an agreement head is adjoined. Finally, theheads are interpreted by the vocabulary items, such as en:[+Agr+pl].

An aspect of DO that is crucial for the following discussions is the as-sumption taken over from DM that agreement has a very special status inthe grammar. While Tense as a lexical head is already present in the syn-tactic module, agreement is only inserted at Chain Interpretation. Moreoveragreement also allows splitting at Head Interpretation5. Thus, as we saw in(1), Georgian subject agreement is sometimes expressed by a single a!x suchas -en, 3pl and sometimes by two a!xes, where person and number is splitas in v- . . . -t .

To account for the complementary distribution of “fused” and “split”agreement, I assume with Noyer (1992) that there is in both cases just oneagreement head at Chain Interpretation (7a), which corresponds to one ortwo vocabulary items at Head Interpretation (7b).

(7) a. [+V]1 [+1 +pl]2 [+V]1 [+3 +pl]2b. v:[+1 +2]2 xedav:[+V]1 t:[+pl]2 xedav:[+V]1 en:[+pl]2

If this analysis is correct, split agreement is of special interest for an ac-count of a!x order, since the respective order of v :[+1] and t :[+pl] cannotfollow from any syntactic principle applying in the synchronic grammar ofGeorgian. Agreement is not a head in syntax proper, and even if it is atChain Interpretation, the vocabulary items at Head Interpretation are justdi"erent parts of this head. Since the features of a head are unordered, allprinciples that regulate the relative order of a!xes corresponding to thesefeatures must be non-syntactic, hence morphological.

Note that in DO there are no words in the traditional sense. However,many morphological processes refer to the Spell-out-domain, a small word-like unit which consist of a lexical head and all string-adjacent functionalcategories from its extended projection (see Trommer 2002a for further dis-cussion) at Head Interpretation.

5

2.2 Optimality Theory and Alignment Constraints

Optimality Theory Prince and Smolensky (1993) is a general formal frame-work based on the assumption of universal violable constraints, which instan-tiate by di"erent rankings grammars of individual languages. OT-constraintsassign to candidates sets of constraint violations. Thus, the constraints from(3) can be restated more formally as the alignment constraints in (8):6

(8) a. [+NUM] ➪ R (Count a constraint violation for each vocabularyitem between a vocabulary item specifying a number feature andthe right edge of the spell-out domain.)

b. L

➪

[+PER] (Count a constraint violation for each vocabularyitem between a vocabulary item specifying a person feature andthe left edge of the spell-out domain.)

Constraints evaluate possible candidates. This process is usually depicted ina tableau like (9) for the Georgian verb form v-xedav-t, we see:

(9) v-xedav-t, ‘we see’

NUM ➪ R L

➪

PER

☞ v-xedav-txedav-v-t *!xedav-t-v *! **v-t-xedav *!

Constraint violations are marked by “*”. Constraint evaluation starts withthe highest-ranked constraint (NUM ➪ R), which eliminates all candidateswhere the number a!x t is not at the right edge (xedav-t-v and v-t-xedav).Elimination of suboptimal candidates is depicted by “!” after the responsibleconstraint violation. The second constraint, L

➪

PER, eliminates now xedav-v-t (1 violation), since this is less harmonic than v-xedav-t which involves noviolation at all.

A major claim of OT is that all constraints are universal and languagesdi"er only by instantiating di"erent rankings of these constraints. In thefollowing sections, I try to show that this also holds for alignment constraintson subject agreement.

6

3 The Language Sample

Split subject agreement and languages where Tense and agreement are bothprefixes (these will become relevant in section 7) are relatively rare. For thisreason, the sample is intended to maximize the number of languages whichexhibit these patterns, since my goal is not to test the frequency of thesepatterns, but di"erent properties of these patterns. On the other hand, Ihave tried to find instances for all patterns in as many genetically diverselanguage families as possible according to the classification of Ruhlen (1987).

A second problem is to di"erentiate subject agreement from similar gram-matical markers such as cliticized pronouns, especially in so-called pronominal-argument languages where agreement a!xes seem to make genuine argu-ments of the verb superfluous.7 In the sample, I have counted all thosemarkers as agreement that encode obligatorily person and/or number fea-tures of the subject and are bound to a verb. This excludes genuinely verbalplural markers which express verbal categories such as iterative or intensityof action, but might be taken as plural agreement in a superficial analysis(cf. Durie 1986). In cases of ergative systems I included the marker agreeingwith the absolutive argument. Finally, I did not consider languages whereSAgr is fused to a high degree with other categories like Tense and Aspect.8

4 The Order of Person and Number Agree-ment

4.1 Methodology

Determining the relative order of subject person and number agreement facesthe problem that person and number markers are often fused, i.e., numbermarkers also specify person values or vice versa. I try to solve the problem byincluding all cases into consideration where one agreement a!x marks onlyone category C1 (person or number) while the other a!x marks the otherone, C2, and possibly also C1, as in (10):

(10) a. A"1 [+PER] A"2 [+NUM]b. A"1 [+PER +NUM] A"2 [+NUM]c. A"1 [+PER] A"2 [+PER +NUM]

7

Linear precedence between a!xes is determined straightforwardly in caseswhere the a!xes cooccur Where they cannot, precedence is determined bythe relative position of the PER/NUM a!xes with respect to a third unit.E.g. the Georgian 3pl su!x -en ([+PER +NUM]) is said to follow the 1sgprefix v- ([+PER]) since the first follows the stem and the latter precedes it,even though this never happens in the same form.

I try to remedy the problem of scarcity of the data by including di"erentpatterns from single or closely related languages. Thus, Quechua is includedtwice in the sample since there are varieties with the order P > N (P(erson)precedes (N)umber) and others with N > P ((N)umber precedes P(erson)).Two patterns are said to be di"erent if they involve di"erent orders or dif-ferent sub-cases of (10).

4.2 Results

In the language sample, there are 58 languages exhibiting splits in per-son/number marking. These show 80 di"erent ordering patterns. 11 of thesepatterns include only prefixes, 28 only su!xes, and 39 are mixed (one is aprefix, and one is a su!x):

(11) Ordering Patterns

10 both prefix 12.5 %30 both su!x 37.5 %40 mixed 50.0 %

What is interesting here is that the mixed cases constitute one half of thepatterns. (12) contains the relative percentages of Person > Number (P >N), and Number > Person (N > P) for all patterns. Note that P > N is thedominant order, no matter which of both a!xes is a prefix or a su!x.

(12) All Patterns

both prefix both su!x mixed all

P > N 9 90.0% 22 73.3% 39 97.5% 70 87.5%N > P 1 10.0% 8 26.7% 1 2.5% 10 12.5%sum 10 30 40 80

8

The same results hold if we look only at the a!xes marking person or number,but not both, i.e., if all patterns involving Person/Number markers (PN) areexcluded:

(13) Only P,N (PN excluded)


P > N 8 88.9% 17 77.3 % 25 100% 50 89.3 %N > P 1 11.1% 5 22.7% 0 0% 6 10.7%sum 9 22 25 56

A possible source of error are person and number a!xes which are alwaysstring-adjacent whenever they cooccur. A complex of such a!xes could ac-tually be a simple a!x. The following table counts only a!xes which are insome instances separated9

(14) Separated A!xes


P > N 2 100.0% 10 66.6% 39 97.5% 51 89.5%N > P 0 0.0% 5 33.3% 1 2.5% 6 10.5%sum 2 15 40 57

Finally, the language sample contains a number of languages which havenumber but no person markers. Here, from 6 languages, the number markeris in 5 languages (83%) a su!x.

4.3 Analysis

The core of my account for the PER-NUM asymmetry (the fact that personagreement tends to precede number agreement) are the two already familiaralignment constraints in (15):

(15) a. L

➪

[+PER] (Person-Agreement is at the left edge.)b. [+NUM] ➪ R (Number-Agreement is at the right edge.)

Considering the possible orders of simple number and person a!xes withrespect to a verbal stem (V), we get the possibilities in (16). The bestcandidate under each ranking is P > V > N, which corresponds closely tothe empirical results, where this is the overall favored ordering of person and

9

number. The orders P > N > V and V > P > N each induce one constraintviolation:

(16) Syntagmatic patterns of P and N

L

➪

PER NUM ➪ R bounded by Data

a. ✌ P > V > N - 24b. ☞ V > P > N * a. 17c. ☞ P > N > V * a. 8d. ☛ V > N > P * ** a.,c. 5e. ☛ N > P > V ** * a.,b. 1f. ✟ N > V > P ** ** a.,b.,c.d.,e. 0

The 4th column in the table states for each candidate by which candidates itis harmonically bounded (Samek-Lodovici and Prince 1999: 2, and referencesthere), i.e., which candidates are more harmonic under all possible constraintrankings. A candidate C harmonically bounds C ! i" C does not induce moreconstraint violations than C ! on any constraint, and C ! induces at least onemore constraint violation than C for at least one constraint. Thus, as faras the alignment constraints are involved, (16c) can never be more harmonicthan (16a). This imposes a natural fitness metric on the candidates whichis indicated here by the symbols ✌ ☞ ☛ ✟. Interestingly, the order we getby this metric corresponds closely to the empirical data from the languagesample. The number of corresponding languages can be found in the fifthcolumn. By the more other candidates a candidate is harmonically bounded,by the less language patterns it is represented in the sample. The mostimportant result is that the pattern P > V > N is the overall favored one,both in the sample and in the order imposed by the constraints.

Taking a di"erent perspective, we can look for the best order of PER/NUMa!xes if this order is already partially determined by other constraints. As-sume that P and N are necessarily prefixes because otherwise some high-ranked third constraint would be violated. This means that we only have theoptions (16c) and (16e), in which case c. always will be preferred:

(17) Syntagmatic patterns of P and N (prefixes)

L

➪


c. ☞ P > N > V * a. 8e. ☛ N > P > V ** * a.,b. 1

10

In a mirror fashion, if some high-ranked constraint forces both P and N to besu!xal, we get (16b,d) as candidates, where the first always out-ranks thelatter:

(18) Syntagmatic patterns of P and N (su!xes)

L

➪


b. ☞ V > P > N * a. 17d. ☛ V > N > P * ** a.,c. 5

If no such restrictions are imposed on P and N, P > V > N is the optimalcandidate. Similar results are obtained for the orders where P and N arepartially fused and for the case where orders are determined paradigmatically.See Trommer (2002a) for details.

So far, the assumption of two alignment constraints makes two predic-tions, namely that P > V > N is the overall favored pattern and that thePER-NUM asymmetry should hold. The latter results from the fact thatfor all constellations (P/N,PN/N,PN/P, paradigmatic or syntagmatic) thecandidates conforming to this generalization always harmonically bound thecandidates violating it. The account makes two further predictions: Lan-guages with number agreement should only have su!xal agreement markerssince L

➪PER is irrelevant here. This is illustrated in (19) where the can-

didate b. is harmonically bounded by a. Again, this conforms with the datapresented in section 4.2.

(19) N only

L

➪


a. ✌ V > N - 5b. ☛ N > V * a. 1

The second prediction is that in languages with fused agreement, i.e., agree-ment where NUM and PER agreement is always expressed by a!xes markingboth, the a!xes should occur consistently prefixally or su!xally. This is truebecause the ranking in (20a) favors su!xes whereas the one in (20b) favorsprefixes. No candidate is harmonically bounded by the other.

11

(20) P and N fused

L

➪

PER NUM ➪ R bounded by

a. ✌NUM V > PN * -b. ✌PER PN > V * -

Assuming that the ranking is essentially arbitrary for fused agreement mark-ing, prefixes and su!xes should be distributed rather evenly in the languagesof the world. This is confirmed empirically by Hawkins and Gilligans (1988:225) who find that certain order preferences for other categories (su!xingover prefixing in certain contexts) do not hold for subject-person marking.In other words, there is no preference for prefixing or su!xing. Since mostinstances of “person marking” in their corpus are person-number marking,this is what we expect. Roughly, the same result is obtained from data pre-sented in Julien (2000: 360): In 93 genera she finds that subject agreementis su!xal while it is prefixal in 85.10 In my data, fused person/number isprefixal in 30 (42,3%) and su!xal in 41 (57,7%) of 71 patterns. However,my data are not completely representative in this respect since certain typesof agreement are over-represented in my sample (cf. section 3).

4.4 Two Case Studies

4.4.1 Muna

In Muna, the number a!x -amu [+pl] follows the stem while the pure personmarkers (o- [+2 -1], no- [+3]) precede it (van den Berg 1989: 51):

(21) o-kala o-kala-amu no-kala do-kalaS2-go S2-go-PL S3-go S3p-go‘you (sg.) go’ ‘you (pl.) go’ ‘he goes’ ‘they go’

This follows from the assumed constraints under any ranking

(22) o-kala-amu, ‘you (pl.) go’

L

➪

PER NUM ➪ R

☞ P > V > NP > N > V *!V > P > N *!N > V > P *!* **

12

That the ranking is indeed L

➪

PER " NUM ➪ R and not the other wayaround can be seen from the fact that fused person/number markers (do-,[+3 +pl]), which are subject to both constraints, share the position of person,not the position of the number marker:

(23) do-kala, ‘they go’

L

➪

PER NUM ➪ R

☞ PN > V *V > PN *!

4.4.2 Georgian

shows the opposite ranking of L

➪

PER and NUM ➪ R as Muna. Whilepure person (v-, 1st person ) and pure number agreement (-t) also occur atthe left and right edge (24) under this ranking, the 3pl marker -en (25) is onthe right of the stem:

(24) v-xedav-t, ‘we see’

NUM ➪ R L

➪

PER

☞ P > V > NP > N > V *!V > P > N *!N > V > P *!* **

(25) xedav-en, ‘they see’

NUM ➪ R L

➪

PER

PN > V *!☞ V > PN *

The case studies for Georgian and Muna show that the proposed alignmentconstraints do not just state crosslinguistic preferences, but – by their di"er-ent rankings in single languages – result in complete accounts of particulargrammars. They also illustrate that the e"ect of alignment cannot be re-duced to the statement that person agreement precedes number agreement

13

since this would not account for the behavior of fused person/number mark-ers, which are positioned di"erently in di"erent languages, but systemati-cally within single languages. For the theory of grammar, replacing subcat-egorization frames of single a!xes by general alignment constraints meanseliminating a major source of language- and even morpheme-specific stipu-lation in favor of a formal device which is independently motivated by itsusefulness to account for such di"erent phenomena as infixation (Prince andSmolensky 1993), stress placement (McCarthy and Prince 1993), clitic or-dering (Grimshaw 2000) and the alignment of syntactic and phonologicalphrases. According to the basic assumptions of OT (Prince and Smolen-sky 1993: 3), the constraint inventory forms the invariable core of universalgrammar. Consequently, the proposed alignment constraints and their pos-sible rankings also constitute a substantial theory of possible a!x orderingsmaking strong falsifiable predictions. The empirical claim of this approachis hence much stronger than that of functional approaches to a!x order,which define only tendencies that cannot be vitiated by negative evidencefrom single languages.11 To maintain this restrictive account we must showthat cases where person agreement apparently follows number agreement donot form real counterevidence against the assumed constraints. This is whatI will try to do in the next section.

4.5 Apparent Counterexamples

Although the PER-NUM asymmetry seems to hold for a wide range of lan-guages, there are counterexamples. I will show that these can be adducedto three factors: 1) Verbal plural markers that do not realize subject agree-ment but are incorporated quantifiers. 2) A!xes that encode other featuresthan person and number, and 3) A!xes which are bound by their context-restrictions to positions which violate alignment constraints.

The first case is exemplified in Jaqaru (Hardman 1966: 53), where theplural marker -rqay precedes person marking:

(26) saynqu-rqay-k-i-waV-PL-TNS-3-TNS

‘they all stand up’

However, Hardman notes that in Jaqaru “plurality is not an inflectional cat-egory” (p. 46). The plural marking both on nouns and verbs is “emphatic”(ibd.) and “the absence of the plural su!x does not imply singular” (p.

14

53). Since “the usual translation is ‘all’ or ‘everyone’ ” (ibd.), it is plausiblethat -rqay is an incorporated quantifier. Similar analyses seem to apply toapparent counterexamples in Navajo (Young and Morgan 1998) and Walapai(Redden 1966).

A second source of the order NUM > PER is due to involvement of a thirdfeature apart from person and number. Thus in Isthmus Zapotec (Pickett1955: 221), person markers follow number markers, but person is generallyfused with gender marking:

(27) ru- unda-ka-beeHABIT-sing-PL-[+3+an]

‘they (an.) sing’

While there are few cases of split gender agreement markers, in Semitic suchmarkers exist and are usually su!xal. This suggests that there is a constraintGEND ➪ R. If this is on the right track, it is plausible that GEND ➪ Ralso can induce violations of the PER-NUM asymmetry. Assuming that inIsthmus Zapotec GEND ➪ R is ranked over NUM ➪ R, and that the prefixoption for subject agreement is not available, we get the observed order:

(28) Ranking of Isthmus Zapotec

GEND ➪ R NUM ➪ R L➪

PER

☞ V [NUM] [PER GEND] * **V [PER GEND] [NUM] *! *

A final source of counterexamples to the PER-NUM asymmetry are a!xesthat are bound to a certain position by a context restriction on the vocabularyitem. One such case is the Ancash Quechua plural su!x -ya (Lakamper andWunderlich 1998) which precedes all person marking (29).

However there is strong evidence that the o"ending order N > P is in-duced by an idiosyncratic context-specification of the number a!x and isnot representative for the general positioning of number agreement in thislanguage. Note first that N > P only occurs in Ancash Quechua while theunmarked (opposite) order holds in Ayacucho Quechua. Even in Ancash,possessor agreement, which is otherwise completely parallel to intransitivesubject agreement, number marking is taken over by another su!x (-kuna)which uniformly follows person markers and thus restates the expected order:

15

(29) Quechua Nominal and Verbal Agreement

Verbs Nounssg pl sg pl

1 -V -ya-V -V -V-kuna2 -nki -ya-nki -yki -yki-kuna3 -n -ya-n -n -n-kuna1+2 -ntsik -ntsik

Finally, the 1st plural inclusive marker -ntsik [+PER +NUM] always occursin final position if it marks object agreement and vacuously if it marks subjectagreement. Again this shows the rightwards tendency of plural marking inthis language.

In DO , the position of -ya can be accounted for by a context restrictionsuch as [+V] on its vocabulary item, which restricts it to the immediateright of a verbal item. Indeed, -ya seems always to occur right-adjacentto the stem or derivational markers12 which can be analyzed as light verbs(cf. Parker 1976). Lakamper and Wunderlich (1998) independently proposesuch a context restriction13 to account for the fact that -ya cannot appearin nominal inflection (cf. (29)). Hence, the context restriction is necessaryfor independent reasons. The account with a context specification also ex-plains the fact that other plural a!xes in Ancash such as the nominal pluralmarker -kuna obey the alignment constraint on plural a!xes. Under lexiconoptimization (Prince and Smolensky 1993), such a context specification willlead to reanalysis whenever it is possible. This is confirmed by the fact thatin three of four Quechua dialects discussed by Lakamper and Wunderlich(1998) -ya is abandoned in favor of a plural marker in the canonical order.

5 The Order of Syntactic Heads

Under the assumptions of DM, split agreement a!xes are a result of mor-phological operations, while interpretable a!xes such as Tense and Aspectnormally have unique correspondents in syntax. The order of these a!xesshould therefore reflect syntactic principles, di"erent from the ones regulat-ing the order of split agreement. In this section, I show that this is indeedthe case, following closely the analysis of Julien (2000).

16

Julien shows that Aspect a!xes appear closer to the stem than Tensea!xes if both are su!xes or both are prefixes. However if the two types ofmarkers appear on di"erent sides of the verb stem, there are clear examplesof the order Tense > Verb > Aspect while the reverse order, Aspect > Verb >Tense, is virtually non-existent. Schematically, we find the pattern in (30):14

(30)

both prefixes Mixed both su!xes

T > A Tense Aspect Verb Tense Verb Aspect *Verb Tense AspectA > T *Aspect Tense Verb *Aspect Verb Tense Verb Aspect Tense

Hence, we find again an ordering restriction (✔Tense > Verb > Aspect but*Aspect > Verb > Tense) that cannot be captured by standard generativeaccounts of a!x order which do not restrict the order of a!xes on oppositesides of the verb.

Julien’s account derives the restrictions depicted in (30) from the as-sumption of a universal phrase structure for functional categories and theindependently motivated antisymmetric framework of Kayne (1994), wherea universally uniform Specifier Head Complement structure and the restric-tion of movement to leftwards movement is assumed. Under this approach,there are four possible situations where Tense, Aspect, and Verb can becomeadjacent, and hence are analyzed as “words”: In the base-generated order,all a!xes stay in the positions where they are generated without movement.This gives the order Tense > Aspect > Verb:

(31) The base-generated order

Tense Aspect Verb

If the verb head-moves up to Asp0, the order Verb > Aspect > Tense results:

17

(32) Head-Movement to Aspect

Tense [Verb Aspect] Verb

!

Further movement of the complex [Verb Aspect ] to Tense0 leads to a standardhead-movement pattern:

(33) Further Movement to Tense

[[Verb Aspect] Tense] [Verb Aspect] Verb

!!

While these options su!ce to derive the three occurring patterns, a fourthscenario is necessary to account for the frequent appearance of su!xes inverb-final languages, which can hardly be the result of moving the head toV, resulting in a sentence-medial position of the verb. Julien (2000: chapter4) shows convincingly that in this language type the dominant pattern ismovement of the complements to the specifiers of functional heads. Thus,in (34) the VP – including all complements – has moved to the specifier ofAsp0. Subsequently, the Aspect Phrase has moved to the specifier of TP:

(34) Phrasal Movement

[[[S O V] Aspect] Tense] [[S O V] Aspect] [S O V]

!!

Note why it is not possible to extend this account to split agreement, wherea!xes are ordered according to the pattern in (35a) while Tense and Aspectreflect (35b):

18

(35) a. Prefixtype1 Prefixtype2 V # V Su!xtype1 Su!xtype2

(type1 = PER, type2 = NUM)b. Prefixtype1 Prefixtype2 V # V Su!xtype2 Su!xtype1

(type1 = Tense, type2 = Aspect)

This indicates strongly that agreement and syntactic heads as Tense andAspect are subject to di"erent principles. However, the antisymmetric andthe alignment-based account are complementary in the sense that togetherthey obviate the need for any idiosyncratic specifications whether a singlea!x is a prefix or a su!x. While this follows from the alignment constraintsin the first case, in the latter it results directly from syntactic movement.Thus the fact that Tense is a prefix in (31) and (32), but a su!x in (33) and(34), results from the movement of the verb to the left of Tense in the latterbut not in the former cases.

In the next two sections, I will investigate how a combination of themovement and the alignment approach can account for restrictions on therelative order of subject agreement and Tense a!xes.

6 The Interaction of Movement and Align-ment

Up to this point, we have identified two grammar modules responsible fordi"erent ordering patterns. The basic question is now whether and how thesemodules interact. In this section, I will investigate this question on the ba-sis of ordering restrictions on subject agreement and Tense inflection. I willassume that the interaction between morphology and syntax in this area isasymmetric: Morphological constraints cannot change the order of elementssuch as Aspect and Tense which is determined by syntax. However, the or-dering of subject agreement is partially determined by syntactic constraints.Since agreement itself is not present in syntax, I attribute this to the factthat agreement is attached to a syntactic category at Chain Interpretationand to a constraint which requires that agreement “reflects” the position ofits host at head interpretation. I will introduce several types of interactionbetween morphology and syntax and show how they can be accounted forunder the proposed assumptions. In section 7, it will be shown that thesame assumptions allow to derive the basic ordering generalizations for thesemarkers that result from my language sample.

19

The simplest conceivable type of interaction is no interaction at all. Thiscan be observed in the Nilo-Saharan language Turkana (Dimmendaal 1983:122,130), as evidenced in the following verb forms:

(36) a. !-a-los-ı b. e-los-e-te c. kı-los-ı3-Past-go-Asp 3-go-Asp-PL S1pl-go-Asp‘he went’ ‘they will go’ ‘we will go’

Tense (a- in (36a)) precedes the verb stem, while Aspect follows it (36a-c).This follows straightforwardly from movement of the verb to the Aspecthead as in (32). The order of agreement a!xes is completely parallel to theone for Muna (see section 4.4): Pure person marking (!- in (36a,b)) andPER/NUM markers (kı- in (36c)) precede all other markers, while the purenumber marker (-te in (36b)) follows all other a!xes. Thus no interactioncan be detected, but crucially Tense and Aspect seem to be part of thealignment domain for agreement. This excludes any account which positsthe application of the morphological constraints in a presyntactic componentof the grammar.

A more interesting type of interaction can be observed in Amharic (Leslau1995: 287,301). In this language there are two basic verbal paradigms, theperfect and the imperfect. In the latter, agreement is again split and the orderof a!xes follows closely the one found in Muna and Turkana. Thus, Personand person/number a!xes such as te- [+2] or enne-:[+1+pl] precede theverb, while the plural su!x -u follows it. However, in the perfect paradigm,all agreement markers are su!xes no matter what features they encode:

(37)

Imperfect Perfect

3. sg. mas ye-saber sabbar-a3. sg. fem te-saber sabbar-acc2. sg. mas te-saber sabbar-h2. sg. fem te-sabr-i sabbar-sh1. sg. e-saber sabbar-hu3. pl. ye-sabr-u sabbar-u2. pl. te-sabr-u sabbar-accuh1. pl. enne-saber sabbar-n

Even if we assume that all perfect markers encode person and number, this

20

cannot be accounted for by invoking the introduced alignment constraints:L

➪

PER favors prefixal agreement, and as is evidenced by the prefix enne-:[+1+pl], L

➪

PER must be ranked above NUM ➪ R in Amharic. But sincethere is a complete person di"erentiation in the perfect paradigm, agreementa!xes must be marked for person.

The analysis I propose is based on the observation that the perfect paradigmis entirely determined by Tense/Aspect, while the position of Imperfect mark-ers can be accounted for entirely by person- and number features. Note firstthat while Leslau uses the aspectual labels Perfect and Imperfect for theseparadigms, his description of their uses (Leslau 1995: 291,311) implies thatthe “imperfect” rather conveys a present tense and the perfect a past tenseinterpretation. Indeed, Cowley et al. (1976: 86) treat these paradigms asdi"erent tenses. Now it is common for languages that present tense remainsunexpressed by a!xes while past tense is expressed by separate a!xes orportmanteau a!xes, as in the following examples from Georgian, where pasttense is expressed by the marker -n or the past/3sg a!x -a while there areno special a!xes for present tense:

(38) a. xedavd-en ‘they see’see-TENSE-S3p

b. xedavd-n-en ‘they saw’see-TENSE-S3p

c. xedav-s ‘he sees’see-S3s

d. xedavd-a ‘he saw’see-TENSE:S3s

Since in Amharic there is no other a!xal reflex of Aspect/Tense, it is plau-sible that the perfect a!xes are actually portmanteau markers which encodepast tense and agreement while the imperfective a!xes are just agreement(the unmarked Tense/Aspect is not overtly realized).15 Assuming that theverb moves to Aspect/Tense, an a!x realizing past should appear as a su!x.Following the assumption that order restrictions established by the syntaxcannot be changed by morphological constraints, markers encoding agree-ment and Tense must reflect the underlying position of the latter, and thisis exactly what we find in Amharic.

A third type of interaction can be found in Island Kiwai (Wurm 1975),where we find again a split of ordering of agreement categories in di"erent

21

tenses, but this time without fusion of Tense and subject agreement. In thepresent and past paradigms person markers are initial and number markers(here: dual) final: 16

(39) Present Near Past Definite Pastn-V-duru-do n-V-do n-V-ru-do1-V-Tns-Du 1-V-Du 1-V-Tns-Du

However, in the future tenses, the dual marker is realized as a prefix followinga prefixal Tense marker:

(40) Indefinite Future Immediate Futureni-du-do-V-ri ni-do-V-ri

1-Tns-Du-V-Tns 1-Du-V-Tns

If we make the plausible assumptions that all present and past forms haveat least one TAM head, and all future forms at least two17, these forms canbe given the structures in (41) and (42):

(41) Kiwai 1st dual form: Present/Past

PRESENT NEAR PAST DEFINITE PASTn- V -duru -do| | |1 Tns1 Du

n- V Ø -do| | |1 Tns1 Du

n- V -ru -do| | |1 Tns1 Du

(42) Kiwai 1st dual forms: Future

INDEFINITE FUTURE IMMEDIATE FUTUREni- du- do- V -ri| | | |1 Tns1 Du Tns2

ni- Ø do- V -ri| | | |1 Tns1 Du Tns2

In an intuitive sense, the dual marker simply follows a certain Tense marker.I suggest to capture this intuition by the following constraint:

(43) REFLECT(AGR): An a!x realizing an agreement category Ashould reflect the position of its host H by

a. being right-adjacent to an a!x realizing H , or byb. occupying the position of H , if H is not realized

22

If we assume that agreement is attached in Island Kiwai to Tense118, this con-

straint will now interact with the already introduced alignment constraintsto give the correct order for all a!xes:

(44) Input: V [+Tense]1 [+1 +du] (PRESENT)

L

➪

PER REFLECT NUM ➪ R

☞ n-V-duru-don-do-V-duru *! **V-duru-n-do *!*do-V-duru-n *!** ***

(45) Input: [+Tense]1 [+1 +du] V [+Tense]2 (INDEF.FUT)

L

➪

PER REFLECT NUM ➪ R

☞ ni-du-do-V-ri **ni-do-du-V-ri *! ***du-ni-do-V-ri *! **ni-du-V-ri-do *! **

Assume now a hypothetical language L, where syntactic constraints on thepositioning of Tense are out-ranked by morphological constraints, hence ir-relevant. L contains a person marker P, a person/number marker PN and aTense marker T. For the ranking REFLECT(AGR) " NUM ➪ R " L

➪

PER, we get the following results. In a form where only P appears, P and Tare prefixes due to the irrelevance of NUM ➪ R:

(46) Influence of morphological constraints on Tense

REFLECT NUM ➪ R L

➪

PER

P T V *!☞ T P V *

V T P **!V P T *! *

However in a form with PN, NUM ➪ R comes crucial and “drags” T intothe su!xal positions where NUM ➪ R and REFLECT can be satisfied:

23

(47) Influence of morphological constraints on Tense

REFLECT NUM ➪ R L

➪

PER

PN T V *! **T PN V *! *

☞ V T PN **V PN T *! * *

L is thus the mirror image of Island Kiwai: While in Kiwai di"erent Tenseheads with di"erent positions induce di"erent positions of one and the sameagreement a!xes, in L di"erent agreement a!xes would enforce di"erentpositions of one and the same Tense a!x.

I hypothesize that such a language is impossible, but if this is true furtherresearch has to reveal why exactly this is the case. Since a!x order in DOas well as in DM is determined after syntax, this does not directly followfrom the derivational priority of syntax. However the priority of syntax overmorphology seems also to hold in other domains such as verb movement(Bobaljik 2001) and feature neutralization (Trommer 2002b).

7 The Relative Order of SAgr and Tense

REFLECT(AGR) and the fact that subject agreement tends to adjoin syn-tactically to Tense (Halle and Marantz 1993: 146)19 let us expect that thereshould be a preference for Tense and agreement occurring on the same sideof the stem because there are more possible rankings that favor “conformity”of Tense and SAgr than rankings disfavoring it:

(48) Tense su!x

Ranking A!xal Status

☞ V > T > A REFLECT AGR " . . . conform☞ V > T > A NUM ➪ R " . . . conform☞ A > V > T L

➪

PER " . . . not conform

24

(49) Tense prefix

Ranking A!xal status

☞ T > A > V REFLECT AGR " . . . conform☞ A > T > V L

➪

PER " . . . conform☞ T > V > A NUM ➪ R " . . . not conform

Indeed, in a majority of languages if Tense and agreement a!xes cooccur,the agreement a!x has the same a!xal status as the Tense a!x. (50) and(51) show the relevant results from my sample and Julien (2000):

(50) A!xal status of SAgr and Tense (my sample)

T su!x T prefix all

Agr conform 48 71.6% 19 70.4% 67 71.3%Agr not conform 19 28.4% 8 29.6% 27 28.7%sum 67 27 94

(51) A!xal status of SAgr and Tense (Julien 2000: 360)

T su!x T prefix all

Agr conform 80 58.4% 23 71.9% 103 60.9%Agr not conform 57 41.6% 9 28.1% 66 39.1%sum 137 32 169

In the following, I will discuss in more detail the possible positions of Tenseand SAgr and show that the assumed constraint set makes the correct pre-dictions for the found ordering preferences. Since my sample is somewhatbiased for the reasons discussed in section 3, I will use in addition the resultsfrom Julien (2000). Note that the references to agreement in the followingrefer to all instances of agreement in my sample (person, number, or personand number fused).

7.1 SAgr and Tense as Prefixes

Assume first that both SAgr and Tense are prefixes. In my sample, thereis a slight preference ($ 60%/40% for the order Tense > Agreement) whilein Julien’s evaluation it is the other way around. Since in this constellation

25

Julien’s and my results di"er, I have additionally evaluated the a!x orderdata she gives in the appendix of her thesis (third column). Here, there isa roughly even distribution of both orders. Probably this divergence is dueto the fact that her criteria to set apart agreement markers from pronouns— which exclude many cases of agreement in polysynthetic languages — arenot reflected in her raw data.

(52) Order of SAgr and Tense as Prefixes

my results Julien’s results Julien’s data (appendix)

T > A 15 62.5% 9 39.1% 17 47.2%A > T 9 37.5% 14 60.9% 19 52.8%sum 24 23 36

Taken together, I interpret these data as evidence that there is no crosslin-guistic evidence for an order preference for Tense/Agreement in prefixal po-sition. This is exactly what is expected given the assumed constraints. Sincethe order Tense > Verb is given by syntax, there are three possible ordersat Head Interpretation: SAgr > Tense > Verb, Tense > SAgr > Verb andTense > Verb >SAgr. The last order results from a high ranked NUM ➪

R and will be discussed in section 7.3. If REFLECT(AGR) is crucially un-dominated, SAgr must appear rightadjacent to Tense; hence we get: Tense> SAgr > Verb. SAgr > Tense > Verb emerges if L

➪

PER dominatesall other crucial constraints since this is the only prefixal order where thisconstraint is unviolated. Putting it in another way: Given Tense > V thereare two relevant constraints that favor prefixal Tense/SAgr. One favors theorder Tense > SAgr and the other SAgr > Tense. If ranking is assumed tobe arbitrary, this should result in an even crosslinguistic distribution of thetwo orders, and this is what is borne out by the data.

7.2 SAgr and Tense as su!xes

While the ordering possibilities of SAgr and Tense as prefixes are rathersymmetric, there is a clear asymmetry in su!xal position. Verb > Tense >SAgr overwhelmingly outranks Verb > SAgr > Tense:

26

(53) Order of SAgr and Tense as Su!xes

my results Julien’s data

T > A 44 84.6% 64 80%A > T 8 15.4% 16 20%sum 52 80

This is again predicted by the assumed set of constraints. Assume thatTense is a su!x. High ranking of L

➪

PER will lead to prefixal agreement(see section 7.3). But if either REFLECT(AGR) or NUM ➪ R are cruciallyundominated, this will result in Verb > Tense > SAgr since the rightmostposition and the position right-adjacent to Tense coincide. Crucially, thereis no constraint that would favor the order Verb > SAgr > Tense.

What remains problematic then is the fact that there are instances of theorder Verb > SAgr > Tense at all. Some of these cases – at least in mysample – are due to special properties of single a!xes. Thus, the Quechuaplural a!x was argued to have the context restriction /[+V] in section4.5 and therefore always precedes Tense. A second class of exceptions arelanguages with the order: Verb > SAgr1 > T > SAgr2. An example is Dumi:

(54) ph k-k -t-aget:up-[+1+pl]-NPast-[-du]

‘we (exc.) get up’ (p. 96)

Note that the o"ending -kA only appears together with a second number a!xthat appears after the Tense marker. In Trommer (2002a) this is related toa constraint that allows only one agreement a!x in a post-tense position.If this outranks REFLECT and NUM ➪ R " L

➪

PER , this leads to theobserved order. Finally, there seem to be languages where SAgr is adjoinedto a non-tense head.20 in which case REFLECT(AGR) is predicted to reflectthe position of its “alternative” host, not of Tense.

7.3 SAgr and Tense on di"erent Sides of the Stem

Siewierska (1993: 68) makes the observation that

(55) ”. . . if SAgr is a su!x so is the Tense marker.”

Siewierska finds that this is true for 91% of the languages from a sample of262 languages. These results are also replicated in other studies

27

(56) Order of SAgr and Tense (Mixed cases)

Siewierska (1993) Julien (2000) my sample

A > V > T 68 88.3% 57 86.4 % 19 70.4%T > V > A 9 11.7% 9 13.6% 8 29.6%sum 77 66 27

This result seems to be problematic, since, according to the proposed con-straints, SAgr > Verb > Tense given su!xal Tense and Tense > Verb > SAgrgiven prefixal Tense are equally probable. In each case, alignment constraintscan be said to outrank REFLECT.

The point however is that su!xal Tense is much more frequent thanprefixal one (Julien 2000: 51 ".). Especially in the patterns involving Tenseand SAgr, su!xal Tense is far more frequent. This is true in my data for 75of 107 patterns (70%) and for Julien’s data in 127 out of 172 (79.6%)

(57) A!xal status of SAgr and Tense (Mixed)

my results Julien’s data

T > A 44 84.6% 64 80%A > T 8 15.4% 16 20%sum 52 80

The distribution of Tense and SAgr in the mixed cases thus seems to followcrucially from the independent distribution of Tense alone.21

8 Word order and the position of agreement

An important part of the literature on a!x order (e.g. Cutler et al. 1985,Bybee et al. 1990, Siewierska and Bakker 1996) tries to relate the position ofa!xes with respect to stems to the basic constituent order of the respectivelanguages. Certainly any account of a!x order should be at least consistentwith the basic findings in this line of research. Siewierska and Bakker (1996:144) give an overview of the results of research on the relation between basicword order types and the position of subject agreement markers in di"erentlanguage samples (S & B = Siewierska and Bakker 1996, H & G = Hawkinsand Gilligan, B & P = Bybee and Perkins, the latter two samples and theStassen sample are documented in Hawkins and Gilligan 1988).22

28

(58) The form of SAgr a!xes relative to basic order(Siewierska and Bakker 1996: 144)

sample su!x prefix both

V3S & B N=80 43 54% 27 34% 10 12%Stassen N=35 24 69% 8 23% 3 9%H & G N=16 12 75% 3 19% 1 6%B & P N=9 4 44% 4 44% 1 12%

V2S & B N=38 11 29% 21 34% 6 16%Stassen N=22 8 36% 11 23% 3 14%H & G N=12 5 42% 5 19% 2 16%B & P N=4 1 25% 2 44% 1 25%

V1S & B N=15 7 47% 6 40% 2 13%Stassen N=7 4 57% 3 43% – –H & G N=10 3 30% 6 60% 1 10%B & P N=5 1 20% 4 80% – –

The results for V1 and V2 languages are rather contradictory. However, thereseems to be a robust trend for agreement a!xes in V-final languages to besu!xal.23 This trend is also corrobated by the data from Julien (2000):24

(59) Position of Agreement markers relative to word order(Julien 2000: 360)

preposed postposed

V-initial 15 15V-medial 62 24V-final 48 73Uncertain 34 127

In the proposed account there is clearly no direct connection between theposition of subject agreement and word order. However there is an indirectlink between the two since agreement is assumed to be attached to Tense,

29

which in turn is ordered by syntactic operations. Thus Tense should workas a type of transmission belt, which mediates the relation of word orderand subject agreement. To test this hypothesis, we first have to look at therelation of the position of Tense and basic word order (Julien 2000). (60)shows the percentages of V-initial, V-medial and V-final languages whichhave prefixal (T > V) and su!xal (V > T) Tense markers.25

(60) Position of Tense markers relative to word order (Julien 2000: 339)

T > V V > T genera

V-initial 24 67% 19 51% 36V-medial 66 65% 67 66% 101V-final 34 25% 127 92% 138

From these data it seems that there is a strong preference for su!xing inV-final languages while both su!xes and prefixes are well-represented inverb-initial and SVO languages. Again data from di"erent samples showconsiderable variation:

(61) The form of Tense a!xes relative to basic order(Hawkins and Gilligan 1988)

sample su!x prefix both

V3Stassen N=35 31 88.6% 2 5.7% 2 5.7%H & G N=15 15 100% 0 0% 0 0%B & P N=14 14 100% 0 0% 0 0%

V2Stassen N=22 14 63.6% 5 22.7% 3 13.6%H & G N=10 5 50% 4 40% 1 10%B & P N=2 0 0% 2 100% 0 0%

V1Stassen N=6 4 66.7% 1 16.7% 1 16.7%H & G N=10 1 10% 4 40% 0 0%B & P N=3 1 33.4% 2 66.6% 0 0%

30

While for SVO and VOS languages there is a rather di"use picture, prefixa-tion being preferred in some and su!xation preferred in other samples, thereis a constant preference for su!xing in OV languages. This contrast betweenVO and OV languages can be accounted for if we assume with Julien (2000)that verb-final languages typically result from the movement of complementsto specifiers, as in (34), while VO languages result from a much wider varietyof syntactic configurations. Now let us assume in a simplifying manner thatTense in (S)V(S)O languages is equally often prefixal and su!xal, while it ismore often su!xal in SOV languages. Recall that the alignment constraintsfor person and number as a whole do not favor su!xes or prefixes since theyact as antagonists. However, COHERENCE lets us expect that agreementshould at least tend to occur at the same side of the verb as Tense. This isclearly borne out for agreement in SOV languages. In SVO/VSO languageswe would expect no preference at all for su!xing or prefixing since there isalso no such preference for Tense and the alignment constraints are neutralin this respect. Given the contradictory tendencies for di"erent samples, thisalso seems to be a reasonable result.

9 Alternative Approaches

Standard generative accounts to a!x order (e.g. Baker 1985, Halle andMarantz 1993, Wunderlich 1996) – while di"ering considerably in basic as-sumptions and details – assume that generalizations on a!x orders exhibita mirror image, i.e., if we find the standard order PrefixType1 PrefixType2

Stem (where the indices Type1 and Type2 stand for a!x types such as per-son agreement, Tense etc.) this implies the standard order Stem Su!xType2

Su!xType1. Moreover, it is assumed that the status of a!xes as prefixes orsu!xes is largely arbitrary, determined by a!x-specific stipulation in lexicalentries. Thus we expect for person and number something like (62):

(62)


P > N *Person Number V Person V Number V Person NumberN > P Number Person V Number V Person *V Number Person

But for person and number a!xes, the data from my sample reveal a com-pletely di"erent picture. First, the order of these categories tends to be

31

Person > Number, no matter whether both are prefixes or both are su!xes.And second, the order of a!xes on di"erent sides of a verb is restricted(✔PER V NUM, but not *NUM V PER), and this seems to be even thestrongest ordering restriction that can be observed for person and numberagreement. Thus we find roughly the distribution in (63):

(63)


P > N Person Number V Person V Number V Person NumberN > P *Number Person V *Number V Person *V Number Person

Moreover, standard generative accounts have no explanation why there shouldbe an ordering asymmetry for fused Person/Number with respect to simplePerson and Number markers.

The situation is slightly di"erent for antisymmetric accounts of a!x order(Kayne 1994, Cinque 1999, Julien 2000) which do predict an asymmetrybetween prefixes and su!xes. Such an account was argued to be correct forTense and Aspect a!xes in section 5. But also for subject agreement thisapproach would incorrectly predict that the ordering preferences for prefixesshould mirror those of su!xes.

The “diachronic syntax hypothesis” o"ers a completely di"erent approachto a!xal status26 assuming that the position of a!xes with respect to stemsshows “a preference for morphemes to be located in the positions of the sepa-rate words which gave rise to them.” (Siewierska and Bakker 1996: 145). Inother words, agreement markers which develop from pronouns (or auxiliaries)in post-verbal position tend to become su!xes, and pronouns/auxiliaries inpreverbal position tend to grammaticalize into prefixes. While the historicaldevelopment of agreement a!xes might ultimately be an important factor ina full understanding of the PER-NUM asymmetry, it o"ers no perspectiveon an adequate synchronic representation of a!x position. Moreover, wehave only few possibilities of historical reconstruction for languages whichshow such a split in their agreement patterns. The only reconstructions forverbal number agreement markers I am aware of are Mithun (1991) for someAmerican Indian languages, and Frajzyngier (1997) for Chadic.27 In bothcases the plural markers seem to have developed from demonstratives and/orpersonal pronouns (which themselves often grammaticalize from demonstra-tives, cf. Lehmann 1995: 37). However, tracing back markers of number

32

agreement to pronouns teaches us nothing about the asymmetry betweenperson and number markers since person markers also derive in the stan-dard case from pronouns (Lehmann 1995: 41). In other words, if both typesof markers derive from the same sources and their position is determinedby their syntactic origins, they should have the same positional properties.However, there might be di"erent historical sources for number agreementmarkers, such as plural words (Dryer 1989), verbal pluractionality markers(Durie 1986, Corbett 2000: chapter 8), distributivity and collective markersor numerals (especially in the case of dual markers).

Interestingly, distributivity and collective markers seem to have ratherdi"erent positional properties than verbal agreement markers. Thus, Navajohas a distributive prefix ıi- which precedes person (and number) marking(Young and Morgan 1998: 62). Similar facts can be observed for Walapai(Trommer 2002a: 151), and also the Jaqaru data discussed in section 4.5show that quantificational elements which are a potential historical sourcefor number agreement markers do not show the order with respect to persona!xes which we expect for number agreement. If “real” number agreementhas developed from such markers its positional properties must be the resultof other factors than merely the position of its grammaticalization source.But even if the preference of number agreement for su!xal position could betraced back to its grammaticalization sources without problems, this wouldnot provide an account for the relative position of person and number whichappear on the same side of the verb.

Siewierska and Bakker (1996) also point to a more general problem withthis type of approach. They observe that object agreement markers tendto follow subject agreement markers no matter on which side of the verbalstem both occur. However, even though Siewierska and Bakker are generallysympathetic to the diachronic syntax approach, they admit that it is not ofmuch use in accounting for ordering patterns of this type since “in view ofthe fact that SAgr and OAgr need not evolve simultaneously, one would notexpect there to be any evident parallels between the order of the subject andobject at clause level and the order of SAgr and OAgr a!xes” (Siewierska andBakker 1996: 149). This argument also carries over to a putative diachronicaccount of the PER-NUM asymmetry: There is no reason to believe thatperson and number agreement markers have become a!xal at the same time,hence no systematic relation between their positions is to be expected.

While the work cited up to this point focuses on a!xal status, Bybee(1985) provides an account for the relative positions of a!xes on the same

33

side of the stem. She argues that the proximity of a!xes to the verbal stemis determined by the relevance of the semantic information of the a!x to theverbal semantics. According to this account, number agreement should tendto appear closer to the stem than person agreement. This leads to roughlythe same empirical predictions as for standard generative accounts whichseem to be empirically incorrect for agreement morphology.

Cutler et al. (1985) advocate a processing-based approach which alsoaddresses the question of a!xal status. They assume that this is governedby a general su!xing preference and the tendency of a!xes to obey the HeadOrdering Principle (HOP), which requires that heads (e.g. the verb in a verbphrase or an inflectional a!x in a word) in a given language consistentlyfollow or precede their complements. This approach is problematic sincesubject agreement does not seem to have a crosslinguistic preference for eithersu!xal or prefixal status, and more generally, because di"erent a!x types(Tense, Aspect, agreement, etc.) show di"erent positional preferences (seealso Julien 2000, Siewierska and Bakker 1996 for further critical discussion).

What is appealing about the accounts of Bybee (1985) and Cutler et al.(1985) is that they try to give a functional motivation of a!x ordering prin-ciples. Could there be a similar argument for grounding the alignment con-straints proposed in this paper? Psycholinguistic evidence shows that thepositions which are targeted by these constraints are the two most promi-nent positions for word recognition: the word initial position is the most cru-cial one while “endings are more salient than middles” (Cutler et al. 1985:743). That person and number are aligned roughly to these positions couldbe seen as a strategy to maximize the recoverability of agreement featuresin language comprehension. In a di"erent way, REFLECT also enhancesrecoverability, since it facilitates the reconstruction of the underlying mor-phosyntactic process, i.e., of the attachment of agreement, to Tense. Thatperson is aligned to the more prominent position could be related to a moregeneral prominence of person over number. Thus, Noyer (1992) finds that indi"erent morphological processes (e.g. a!x choice for templatic positions)person features have priority over number features and Corbett (2000: 278)states that in paradigms of pronouns and agreement a!xes “number usu-ally depends on person, but person may depend on number”.28 Data fromSiewierska and Bakker (1996: 122) show that for their language sample pureperson agreement markers are significantly more frequent (21 = 11.7% of 180markers) than pure number markers (12 = 6.7% of 180 markers) and alsothat languages with person agreement (93% of 237) are slightly more frequent

34

than those with number agreement (88% of 237).29 While these remarks arehighly speculative, it seems plausible that the account in terms of alignmentconstraints is not only empirically superior to alternative approaches, butcan also be motivated on functional grounds.

10 Summary

In this paper I have shown that the interaction of morphology and syntax inan OT grammar accounts for a wide range of facts about the order of subjectagreement a!xes, both crosslinguistically and in the grammar of single lan-guages. It remains to be seen how this type of account extends to other a!xtypes. Trommer (2002a) shows that the ordering of fused subject/objectagreement also follows an alignment pattern and discusses some tentativeresults on object agreement pointing in the same direction. Hyman (2001)shows that a!xes involved in Bantu grammatical-function changing processes(causative, passive, etc.) also show the interplay of syntactic/semantic con-straints with purely morphological factors.30 Interestingly, there is indepen-dent morphological evidence that at least voice a!xes form a natural classwith agreement and are inserted “late” after syntax (Embick 1998). Thisgives further support to the claim that the asymmetry between Tense/Aspectand Agreement is due to the more principled di"erence between syntactic andmorphological a!xes.

Appendix: The Language Survey

The following table lists the languages from my language sample. The secondcolumn contains the phyla according to Ruhlen (1987), the third column thelinguistic macro-area according to Dryer (1992), and the fourth column thesource of the data:31

Language Phylum Macro-Areas SourceAcoma Keresiouan N.AMERICA Miller (1965)Ainu Korean-Japanese EURASIA Shibatani (1990)Akan Kwa AFRICA Campbell (1991)Albanian Indoeuropean EURASIA Buchholz and Fiedler (1987)Aleut Eskimo-Aleut EURASIA Bergsland (1994)Amharic Semitic AFRICA Leslau (1995)Anywa Nilotic AFRICA Reh (1993)

35

Apalai Carib S.AMERICA Koehn and Koehn (1986)Arabic Semitic AFRICA Ouhalla (1991)Axininca Equatorial S.AMERICA Payne (1981)Jaqaru Aymaran S.AMERICA Hardman (2000)Azerbajianian S.Turkic EURASIA Schonig (1998)Basque Isolate EURASIA Arregi (1999)Beja Cushitic AFRICA Hudson (1976)Berber Berber AFRICA Noyer (1992)Brahui NW.Dravidian EURASIA Elfenbein (1998)Breton Celtic EURASIA Press (1986)Cayuvava Equatorial S.AMERICA Key (1967)Chamorro W.M.-Polynes. SE.ASIA/OC. Chung (1982)Chinook Penutian N.AMERICA Andersen (1977)Choktaw Penutian N.AMERICA Broadwell (2000)Chukchi Ch.-Kamchatkan EURASIA Krause (1976)Chuvash Bolgar EURASIA Johanson and Csato (1998)Didinga Saharan AFRICA Bryan and Tucker (1966)Dumi Tibetic SE.ASIA/OC. van Driem (1993)Dyola NW.Atlantic AFRICA Givon (1975)Elamite Elamo-Dravidian EURASIA Reiner (1969)Evenki Tungus EURASIA Nedyalkov (1994)Fula W.Atlantic AFRICA Arnott (1970)Fur Fur AFRICA Jakobi (1972)Gahuku Indo-Pacific AUSTR./NG. Foley (1986)Georgian S.Caucasian EURASIA Aronson (1982)German Germanic EURASIAHixkaryana Carib S.AMERICA Derbyshire (1979)Huave Penutian N.AMERICA Stairs and Hollenbach (1969)Inuktitut Eskimo EURASIA Mallon (1991)Jacaltec Otomanguean N.AMERICA Day (1973)Juang Austroasiatic SE.ASIA/OC. Mahapatra (1976)Jyarong Tibetic SE.ASIA/OC. DeLancey (1985)Kalasala Nuristani EURASIA Degener (1998)Kalmyk Mongolian EURASIA Campbell (1991)Kanuri Saharan AFRICA Cy"er (1992)Ket Isolate EURASIA Noyer (1992)Khanty Ugric EURASIA Abondolo (1998)Kilivila CE.M.-Polynes. SE.ASIA/OC. Senft (1986)

36

Kiwai Trans-Fly AUSTR./NG. Wurm (1975)Kobon Trans-NewGuinea AUSTR./NG. Davies (1989)Lenakel CE.M-Polynes. SE.ASIA/OC. Tryon (1973)Lithuanian Balto-Slavic EURASIA Eckert et al. (1994)Macushi Carib S.AMERICA Abbott (1991)Mansi Ugric EURASIA Keresztes (1998)Mapudungun Andean SE.ASIA/OC. Grimes (1985)Marathi Indic EURASIA Pandharipande (1997)Maricopa Hokan N.AMERICA Gordon (1986)Maung Yiwaidjan AUSTR./NG. Donohue (1998)Mekeo Mek AUSTR./NG. Jones (1998)Menomini Algonquian N.AMERICA Bloomfield (1962)Mohawk Keresiouan N.AMERICA Bonvillain (1973)Muna W.M.-Polynes. SE.ASIA/OC. van den Berg (1989)Nahuatl Uto-Aztecan NAMERICA Andrews (1975)Nandi Nilotic AFRICA Creider and Creider (1989)Nenets Samoyed EURASIA Salminen (1998)Nimboran Nimboran AUSTR./NG. Inkelas (1993)Nocte Tibetic SE.ASIA/OC. Gupta (1971)Nungali Djamindjungan AUSTRIA Hoddinott and Kofod (1976)Nunggubuyu Australian AUSTR./NG. Heath (1984)Nyangumarda Pama-Nyungan AUSTR./NG. Hoard and O’Grady (1976)Piro Equatorial S.AMERICA Matteson (1965)Portuguese Italic EURASIA Iliescu and Mourin (1991)Quechua Quechuan S.AMERICA Lakamper and Wunderlich (1998)Quileute Chimakuan N.AMERICA Andrade (1922)Saamic Finnic EURASIA Sammallahti (1998)Sanskrit Indic EURASIA Bucknell (1994)Seri Hokan N.AMERICA Marlett (1990)Shinasha Omotic AFRICA Rottland (1990)Somali Cushitic AFRICA El-Solami-Mewis (1987)Sora Austroasiatic EURASIA Baker (1985)Straits Salish N.AMERICA Jelinek and Demers (1994)Swahili Bantu AFRICA Vitale (1981)Tama Nilotic AFRICA Bryan and Tucker (1966)Tamil S.Dravidian EURASIA Annamalai and Steever (1998)Teda Saharan AFRICA Bryan and Tucker (1966)Timucua Paezan S.AMERICA Granberry (1993)

37

S.Tiwa Tanoan N.AMERICA Rosen (1990)Turkana Nilotic SE.ASIA/OC. Dimmendaal (1983)Tzotzil Penutian N.AMERICA Aissen (1987)Ubykh N.Caucasian EURASIA Campbell (1991)Udmurt Finnic EURASIA Salminen (1998)Urubu-Kaapor Tupi-Guarani S.AMERICA Kakumasu (1991)Walapai Hokan N.AMERICA Redden (1966)Wardaman Gunwinyguan AUSTR./NG. Merlan (1994)Warlpiri Pama-Nyungan AUSTR./NG. Hale (1973)WesternDesert Pama-Nyungan AUSTR./NG. Noyer (1992)Winnebago Keresiouan N.AMERICA Greenberg (1988)Yakut N.Turkic EURASIA Stachowski and Menz (1998)Yimas Nor-Pondo AUSTR./NG. Foley (1991)Yukaghir Uralic-Yukaghir EURASIA Campbell (1991)Yurok Ritwan N.AMERICA Robins (1958)Zapotec Otomanguean N.AMERICA Pickett (1955)

Notes

1 All Georgian data are from Carmack (1997: 315,321). Abbreviations used in theglosses and the text: A! = a"x, A(gr) = agreement, Asp = Aspect, D = determiner, du= dual, exc. = exclusive (plural), fem, = feminine gender, Fut = future tense, GEND =gender, Habit = habitual aspect, IP = Inflectional Phrase, mas = masculine gender, Nom= nominative, N(UM) = number agreement, NPast = non-past tense, P(ER) = personagreement, pl = plural, PN = person-number agreement, S = Subject, S1s = Subject firstperson singular, S1p = Subject first person plural, SAgr = subject agreement, T(ns) =Tense, TP = Tense phrase, VP = verbal phrase

2See section 2.2 for more discussion of alignment constraints.3 Wurm gives only schematic paradigms and does not discuss the meanings and uses

of the various tenses. Therefore, I do not attempt to gloss his data.4 Note that this is not agreement but just the representation of a chain according to

the copy theory of movement (Chomsky 1995).5Split agreement corresponds closely to the notion of discontinuous bleeding from Noyer

(1992).6The alignment constraints that I use here are formally much simpler than “Generalized

Alignment” (GA) as proposed by McCarthy and Prince (1993). This is partly due toexpository convenience, and partly to the assumption that a"x order constraints requireonly a small subset of the formal power embodied by GA. All of the following analyses

38

can be straightforwardly translated into the richer GA format.7Even in languages where NPs are arguably not arguments this does not mean that the

agreement a"xes are pronouns. Thus Baker (1996) argues convincingly for Mohawk thatin such a case the agreement markers identify empty pronominals and are not themselvesincorporated pronouns.

8But see the discussion of the Amharic perfect paradigm in section 6.9I also checked whether areal factors influence the results. To this aim, I checked the

di!erent ordering distribution in the six large geographical areas of the world accordingto Dryer (1992). According to my data, the preference for the order P > N holds in all ofthese areas. See Trommer (2002a) for the detailed results.

10I considered only the cases where SAgr is an a"x bound to the verb while Julien alsoconsiders other realizations of SAgr. Taking the full range of data into consideration doesnot change the results. See footnote 21 for a possible explanation for the slight preferenceof su"xal SAgr.

11See section 9 for discussion of functional and historical approaches to a"x order.12E.g. causative a"xes.13With the di!erence that a context restriction under their premises does not imply

adjacency.14 Julien’s claim seems to concern only syntactic heads, hence inflectional Aspect and

Tense and not cases where Aspect is – even though productively – derivational, such asSlavic aspectual prefixes (Spencer 1991: 195, Filip 2002).

15An anonymous reviewer claims that all person/number markers are portmanteau af-fixes expressing agreement and aspect. While this might be an inevitable analysis in alexicalist theory, it is not in Distributed Morphology, where vocabulary items are under-specified for the syntactic features they realize. I also assume that the di!erent stemshapes are due to morphophonological rules, which is supported by the fact that they arelargely governed by lexical classes.

16See footnote 3 on the presentation of the Kiwai data.17The first assumption reduces to the claim that finite sentences obligatorily contain

Tense (Chomsky 1995: ch.4). While the semantics of the future forms is not clear fromWurm’s description, we would expect that they di!er on the value of a morphosyntacticfeature F , which is represented as the head du- in the indefinite future. That the corre-sponding value for the immediate future is also projected in the same syntactic position isnatural in a model where the semantic content and scope of syntactic heads are reflectedin universally fixed base configurations (see section 5).

18That Agreement adjoins to Tense is not an idiosyncratic assumption about Kiwai, butis also reflected in the crosslinguistic positioning of Tense and Agreement. See section 7for discussion.

19The structural proximity of Tense and subject agreement is also reflected by the factthat these categories often fuse into portmanteau markers, as discussed for Georgian andAmharic in section 6.

39

20This conclusion is drawn for independent reasons in Julien (2000: 214).21 The preference for su"xal Tense seems to be due to the fact that languages with

basic SOV order have a preference for su"xal Tense while SVO and verb-initial languageshave no clear preference with regard to afixal status. See section 8 for further discussionof this topic. The preference for su"xal Tense is also a possible explanation for the slightpreference for su"xal SAgr in the data of Julien (cf. section 4), since REFLECT favorsSAgr and Tense on the same side of the verb.

22A further extensive study on the position of agreement a"xes is Bybee et al. (1990).Bybee et al. (1990) investigate subject and object agreement as a whole without givingdi!erent figures for both types of agreement. Since my study is restricted to subjectagreement, and Siewierska and Bakker (1996) find significant di!erences between theirordering properties this study is not taken into account here.

23This is roughly in line with the Head Ordering Principle (HOP) proposed by Cutleret al. (1985) and Hawkins and Gilligan (1988). However the HOP is problematic for othertypes of a"xes. See Siewierska and Bakker (1996) for critical discussion.

24No percentages are provided since Julien neither gives them nor the total number ofgenera considered for each word order type, which would allow to compute them.

25Note that the occurence of both, preposed and postposed markers in a single language(genus), are possible. Therefore the rows add to more than 100%.

26Bybee et al. (1990: 3) call the same assumption “fossilized syntax hypothesis”.27See Corbett (2000: 266-68) for an overview.28For example, he finds that ”if number is not available for all persons, then it will be

found first of all in the first person, then in the second, and in the third only if in bothfirst and second also” (Corbett 2000: 277).

29Languages with person and number agreement fall under both categories.30In contrast to my approach, Hyman formulates morphological constraints by means

of a (violable) template.31 CE. = Central-East, N. = North, NW. = North-West, S. = South , SE. = South-East,

W. = West, OC. = OCEANIA, AUSTR./NG. = Australia/New Guinea, M.-Polynes. =Malayo-Polynesian, Ch.-Kamchatkan = Chukotko-Kamchatkan

ReferencesAbbott, M. 1991. “Macushi”. In D. C. Derbyshire and G. K. Pullum (eds.), Handbook of

Amazonian Languages, volume 2. Berlin: Mouton de Gruyter, 23–160.Abondolo, D. 1998. “Khanty”. In D. Abondolo (ed.), The Uralic Languages. London:

Routledge, 358–386.Aissen, J. L. 1987. Tzotzil Clause Structure. Dordrecht: Reidel.Andersen, S. R. 1977. “On Mechanisms by which Languages Become Ergative”. In C. Li

(ed.), Mechanisms of Syntactic Change, volume 2. Austin: University of Texas Press,317–364.

40

Andrade, M. J. 1922. “Quileute”. In F. Boas (ed.), Handbook of American Indian Lan-guages, volume 2. Smithsonian Institution, Bureau of American Ethnology, Bulletin 40.Washington: Government Printing O"ce, 149–292.

Andrews, J. R. 1975. Introduction to Classical Nahuatl. Austin: University of Texas Press.Annamalai, E. and S. B. Steever. 1998. “Modern Tamil”. In S. B. Steever (ed.), The

Dravidian Languages. London: Routledge, 100–128.Arnott, D. W. 1970. The Nominal and Verbal Systems of Fula. Oxford: Clarendon Press.Aronson, H. I. 1982. Georgian: A Reading Grammar. Bloomington: Slavica.Arregi, K. 1999. “Person and Number Inflection in Basque”. In V. Lin, C. Krause,

B. Bruening and K. Arregi (eds.), Papers on Morphology and Syntax: Cycle Two,volume 34 of MIT Working Papers in Linguistics. Cambridge Mass: MITWPL, 229–264.

Baker, M. C. 1985. “The Mirror Principle and Morphosyntactic Explanation”. LinguisticInquiry 16, 537–576.

Baker, M. C. 1996. The Polysynthesis Parameter. Oxford: Oxford University Press.Bergsland, K. 1994. “Aleut Tenses and Aspects”. In C. Bache (ed.), Tense, Aspect and

Action: Empirical and Theoretical Contributions to Language Typology. Berlin: Moutonde Gruyter, 323–369.

Bloomfield, L. 1962. The Menomini Language. New Haven: Yale University Press.Bobaljik, J. D. 2001. “Realizing Germanic Inflection: Why Morphology does not Drive

Syntax”. Ms., McGill University.Bonvillain, N. 1973. A Grammar of Akwesasne Mohawk. Ottawa: National Museum of

Man.Broadwell, G. A. 2000. “Choktaw”. Ms., University at Albany, State University of New

York.Bryan, M. A. and A. Tucker. 1966. Linguistic Analyses: The Non-Bantu Languages of

North-Eastern Africa. Oxford: Oxford University Press.Buchholz, O. and W. Fiedler. 1987. Albanische Grammatik. Leipzig: VEB Verlag Enzyk-

lopadie.Bucknell, R. S. 1994. Sanskrit Manual. Delhi: Motilal Banarsidass.Bybee, J. L. 1985. Morphology. Amsterdam: John Benjamins.Bybee, J. L., W. Pagliuca and R. D. Perkins. 1990. “On the Asymmetries in the A"xation

of Grammatical Material”. In W. Croft, K. Denning and S. Kemmer (eds.), Studies inTypology and Diachrony. Amsterdam: John Benjamins, 1–42.

Campbell, G. L. 1991. Compendium of the World’s Languages, volume 1. London: Rout-ledge.

Carmack, S. 1997. “Blocking in Georgian Verb Morphology”. Language 73, 314–338.Chomsky, N. 1995. The Minimalist Program. Cambridge Mass: MIT Press.Chung, S. 1982. “Unbounded Dependencies in Chamorro Grammar”. Linguistic Inquiry

13, 39–77.Cinque, G. 1999. Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford:

Oxford University Press.Corbett, G. G. 2000. Number. Cambridge Textbooks in Linguistics. Cambridge: Cam-

bridge University Press.

41

Cowley, R., M. L. Bender and C. A. Ferguson. 1976. “The Amharic Language”. In M. L.Bender, J. Bowen, R. Cooper and C. A. Ferguson (eds.), Language in Ethiopia. Oxford:Oxford University Press, 63–76.

Creider, C. A. and J. T. Creider. 1989. A Grammar of Nandi. Hamburg: Helmut BuskeVerlag.

Cutler, A., J. A. Hawkins and G. Gilligan. 1985. “The Su"xing Preference: A ProcessingExplanation”. Linguistics 23 , 723–758.

Cy!er, N. 1992. We Learn Kanuri. Koln: Rudiger Koppe Verlag.Davies, J. 1989. Kobon. London: Routledge.Day, C. 1973. The Jacaltec Language. Bloomington: Indiana University Publications.Degener, A. 1998. Die Sprache von Nisheygram im afghanischen Hindukush. Wiesbaden:

Harrassowitz Verlag.DeLancey, S. 1985. “An Interpretation of Split Ergativity and Related Patterns”. Language

51, 626–657.Derbyshire, D. C. 1979. Hixkaryana. Amsterdam: North Holland Publishing Company.Dimmendaal, G. J. 1983. The Turkana Language. Dordrecht: Foris.Donohue, M. 1998. “A Note on Verbal Agreement in Maung”. Ms., University of Manch-

ester.Dryer, M. S. 1989. “Plural Words”. Linguistics 27, 865–95.Dryer, M. S. 1992. “The Greenbergian Word Order Correlations”. Language 68, 81–138.Durie, M. 1986. “The Grammaticization of Number as a Verbal Category”. In Proceedings

of the 12th Annual Meeting of the Berkeley Linguistics Society, 355–370.Eckert, R., Bukeviciute, E.-J. and F. Hinze. 1994. Die baltischen Sprachen. Leipzig:

Langenscheidt Verlag Enzyklopadie.El-Solami-Mewis, C. 1987. Lehrbuch des Somali. Leipzig: VEB Verlag Enzyklopadie.Elfenbein, J. 1998. “Brahui”. In S. B. Steever (ed.), The Dravidian Languages . London:

Routledge, 388–414.Embick, D. 1998. “Voice Systems and the Syntax/Morphology Interface”. In H. Harley

(ed.), Papers from the UPenn/MIT Roundtable on Argument Structure and Aspect,volume 34 of MIT Working Papers in Linguistics. Cambridge Mass: MITWPL, 41–72.

Filip, H. 2002. “Prefixes and the Delimitation of Events”. Ms., Northwestern University.Foley, W. A. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge

University Press.Foley, W. A. 1991. The Yimas Language of New Guinea. Stanford: Stanford University

Press.Frajzyngier, Z. 1997. “Grammaticalization of Number: from Demonstratives to Nominal

and Verbal Plural”. Journal of Linguistic Typology 1, 193–242.Givon, T. 1975. “Serial Verbs and Syntactic Change: Niger-Congo”. In C. Li (ed.), Word

Order and Word Order Change. Austin: University of Texas Press, 47–112.Gordon, L. 1986. Maricopa Morphology and Syntax. Berkeley: University of California

Press.Granberry, J. 1993. A Grammar and Dictionary of the Timucua Language. Tuscaloosa:

University of Alabama Press.

42

Greenberg, J. H. 1988. “The First Person Inclusive Dual as an Ambiguous Category”. InJ. H. Greenberg (ed.), Studies in Language. Amsterdam: John Benjamins, 1–18.

Grimes, J. E. 1985. “Topic Inflection in Mapudungun Verbs”. IJAL 51, 141–163.Grimshaw, J. 2000. “Optimal Clitic Positions and the Lexicon in Romance Clitic Systems”.

In G. Legendre, J. Grimshaw and S. Vikner (eds.), OT Syntax. Cambridge Mass: MITPress, 205–240.

Gupta, D. 1971. An Introduction to the Nocte Language. Shillong: North-East FrontierAgency.

Hale, K. 1973. “Person Marking in Walbiri”. In S. Anderson and P. Kiparsky (eds.), AFestschrift for Morris Halle. New York: Holt, Rinehart and Winston, 308–344.

Halle, M. and A. Marantz. 1993. “Distributed Morphology and the Pieces of Inflection”.In K. Hale and S. J. Keyser (eds.), The View from Building 20. Cambridge Mass: MITPress, 111–176.

Hardman, M. J. 1966. Jaqaru: Outline of Phonological and Morphological Structure. TheHague: Mouton.

Hardman, M. J. 2000. Jaqaru. Munchen: Lincom Europa.Hawkins, J. and G. Gilligans. 1988. “Left-Right Asymmetries in Morphological Univer-

sals”. In J. Hawkins and H. Holmback (eds.), Papers in Universal Grammar. Lingua(special issue), 221–244.

Hawkins, J. A. and G. Gilligan. 1988. “Prefixing and Su"xing Universals in Relation toBasic Word Order”. Lingua 74, 219–259.

Heath, J. 1984. Functional Grammar of Nunggubuyu. Canberra: Australian Institute ofAboriginal Studies.

Hoard, J. E. and G. N. O’Grady. 1976. “Nyangumarda Phonology: A Preliminary Report”.In R. M. Dixon (ed.), Grammatical Categories in Australian Languages. Canberra: Aus-tralian Institute of Aboriginal Studies, 51–77.

Hoddinott, W. G. and F. M. Kofod. 1976. “Djamidjungan”. In R. M. Dixon (ed.),Grammatical Categories in Australian Languages. Canberra: Humanities Press, 397–401.

Hudson, R. A. 1976. “Beja”. In M. L. Bender (ed.), The Non-Semitic Languages ofEthiopia, number 5 in Occasional Papers Series. East Lansing: African Studies Center,Michigan State University, 97–132.

Hyman, L. M. 2001. “Templatic and/or Compositional Morphology: Su"x Ordering inBantu”. Paper presented at the Bay Area Typology Workshop, March 17, 2001.

Iliescu, M. and L. Mourin. 1991. Typologie de la Morphologie Verbale Romance. Innsbruck:Verlag des Instituts fur Sprachwissenschaft der Universitat Innsbruck.

Inkelas, S. 1993. “Nimboran Position Class Morphology”. Natural Language and LinguisticTheory 11, 559–624.

Jakobi, A. 1972. A Fur Grammar. The Hague: Mouton.Jelinek, E. and R. A. Demers. 1994. “Predicates and Pronominal Arguments in Straits

Salish”. Language 16, 697–735.Johanson, L. and E. A. Csato. 1998. “Chuvash”. In S. B. Steever (ed.), The Turkic

Languages . London: Routledge, 434–452.

43

Jones, A. 1998. Towards a Lexicogrammar of Mekeo. Canberra: Australian NationalUniversity Research School of Pacific and Asian Studies.

Julien, M. 2000. Syntactic Heads and Word-Formation: A Study of Verbal Inflection. PhDthesis, University of Tromsø.

Kakumasu, J. 1991. “Urubu-Kaapor”. In D. C. Derbyshire and G. K. Pullum (eds.),Handbook of Amazonian Languages, volume 1. Berlin: Mouton de Gruyter, 326–406.

Kayne, R. S. 1994. The Antisymmetry of Syntax. Cambridge Mass: MIT Press.Keresztes, L. 1998. “Mansi”. In D. Abondolo (ed.), The Uralic Languages. London:

Routledge, 387–427.Key, H. H. 1967. Morphology of Cayuvava. The Hague: Mouton.Koehn, E. and S. Koehn. 1986. “Apalai”. In D. C. Derbyshire and G. K. Pullum (eds.),

Handbook of Amazonian Languages, volume 1. Berlin: Mouton de Gruyter, 33–127.Krause, S. R. 1976. Topics in Chukchee Phonology and Morphology. PhD thesis, University

of Illinois at Urbana-Champaign.Lakamper, R. and D. Wunderlich. 1998. “Person Marking in Quechua – a Constraint-based

Minimalist Analysis”. Lingua 105, 113–148.Lehmann, C. 1995. Thoughts on Grammaticalization, volume 1 of Lincom Studies in

Theoretical Linguistics. Munchen: Lincom Europa.Leslau, W. 1995. Reference Grammar of Amharic. Wiesbaden: Harrassowitz Verlag.Mahapatra, B. P. 1976. “Comparative Notes on Juang and Kharia Finite Verbs”. In

P. N. Jenner, L. C. Thompson and S. Starosta (eds.), Austroasiatic Studies, volume 2.Honolulu: University Press of Hawaii, 801–814.

Mallon, M. 1991. Introductory Inuktitut: Reference Grammar. Montreal: Arctic College -McGill University Inuktitut Text Project.

Marlett, S. A. 1990. “Person and Number Inflection in Seri”. IJAL 56, 503–541.Matteson, E. 1965. The Piro (Arawakan) Language. Berkeley: University of California

Press.McCarthy, J. and A. Prince. 1993. “Generalized Alignment”. In G. Booij and J. van Marle

(eds.), Yearbook of Morphology 1993. Dordrecht: Kluwer, 79–153.McCarthy, J. and A. Prince. 1994. “The Emergence of the Unmarked: Optimality in

Prosodic Morphology”. In Proceedings of the North East Linguistic Society 24, 333–379.

Merlan, F. C. 1994. A Grammar of Wardaman. Berlin: Mouton de Gruyter.Miller, W. R. 1965. Acoma Grammar and Texts. Berkeley: University of California Press.Mithun, M. 1991. “The Development of Bound Pronominal Paradigms”. In W. P. Lehmann

and H.-J. J. Hewitt (eds.), Language Typology 1988: Typological Models in Reconstruc-tion, volume 81 of Current Issues in Linguistic Theory. Amsterdam: John Benjamins,85–104.

Nedyalkov, I. 1994. “Evenki”. In P. Kahrel and R. van den Berg (eds.), Typological Studiesin Negation. Amsterdam: John Benjamins, 1–34.

Noyer, R. R. 1992. Features, Positions and A!xes in Autonomous Morphological Struc-ture. PhD thesis, MIT.

Ouhalla, J. 1991. Functional Categories and Parametric Variation. London: Routledge.Pandharipande, R. V. 1997. Marathi. London: Routledge.

44

Parker, G. 1976. Gramatica Quechua: Ancash-Huailas. Lima: Ministerio de Educacion,Instituto de Estudios Peruanos.

Payne, D. L. 1981. The Phonology and Morphology of Axininca Campa. Dallas: TheSummer Institute of Linguistics and the University of Texas at Arlington.

Pickett, V. 1955. “Isthmus Zapotec Verb Analysis II”. IJAL 21, 217–232.Press, I. 1986. A Grammar of Modern Breton. Berlin: Mouton de Gruyter.Prince, A. and P. Smolensky. 1993. Optimality Theory: Constraint Interaction in Gen-

erative Grammar. Technical Reports of the Rutgers University Center of CognitiveScience. RUCCS TR-2.

Redden, J. E. 1966. “Walapai II: Morphology”. IJAL 32, 141–163.Reh, M. 1993. Anywa Language. Koln: Rudiger Koppe Verlag.Reiner, E. 1969. “The Elamite Language”. In B. Spuler (ed.), Altkleinasiatische Sprachen,

Handbuch der Orientalistik. Leiden: E. J. Brill, 54–116.Robins, R. H. 1958. The Yurok Language. Berkeley: University of California Press.Rosen, C. 1990. “Rethinking Southern Tiwa: The Geometry of a Triple-Agreement lan-

guage”. Language 66, 669–709.Rottland, F. 1990. “A Sketch of Shinasha Morphology”. In R. J. Hayward (ed.), Omotic

Language Studies . London: School of Oriental and African Studies, University of Lon-don, 185–209.

Ruhlen, M. 1987. A Guide to the World’s Languages: Classification, volume 1. StanfordUniversity Press.

Salminen, T. 1998. “Nenets”. In D. Abondolo (ed.), The Uralic Languages. London:Routledge, 516–547.

Samek-Lodovici, V. and A. Prince. 1999. Optima. Technical Reports of the RutgersUniversity Center of Cognitive Science. RUCCS TR-57.

Sammallahti, P. 1998. “Saamic”. In D. Abondolo (ed.), The Uralic Languages. London:Routledge, 387–427.

Schonig, C. 1998. “Azerbaijanian”. In S. B. Steever (ed.), The Turkic Languages . London:Routledge, 248–260.

Selkirk, E. 1995. “The Prosodic Structure of Function Words”. In Papers in OptimalityTheory, volume 18 of University of Massachusetts Occasional Papers. Amherst: Uni-versity of Massachusetts, 439–469.

Senft, G. 1986. Kilivila: The Language of the Trobriand Islanders. Berlin: Mouton deGruyter.

Shibatani, M. 1990. The Languages of Japan. Cambridge: Cambridge University Press.Siewierska, A. 1993. “The Relationship between A"x and Main Clause Constituent Or-

der”. In C. Wilder and D. Cavar (eds.), Word Order Variation, Verb Movement andEconomy Principles. Frankfurt a. M.: Johann Wolfgang Goethe Universitat, 63–75.

Siewierska, A. and D. Bakker. 1996. “The Distribution of Subject and Object Agreementand Word Order Type”. Studies in Language 20, 115–161.

Spencer, A. 1991. Morphological Theory. Oxford: Blackwell.Stachowski, M. and A. Menz. 1998. “Yakut”. In S. B. Steever (ed.), The Turkic Languages.

London: Routledge, 417–433.Stairs, E. F. and B. E. Hollenbach. 1969. “Huave Verb Morphology”. IJAL , 38–53.

45

Trommer, J. 2002a. Distributed Optimality. PhD thesis, University of Potsdam.Trommer, J. 2002b. “Modularity in OT-Morphosyntax”. In G. Fanselow and C. Fery

(eds.), Resolving Conflicts in Grammar: Optimality Theory in Syntax, Morphology andPhonology. Special Issue 11 of Linguistische Berichte.

Tryon, D. T. 1973. “Linguistic Subgrouping in the New Hebridies: A Preliminary Ap-proach”. Oceanic Linguistics 12, 303–351.

van den Berg, R. 1989. A Grammar of the Muna Language. Dordrecht: Foris.van Driem, G. 1993. A Grammar of Dumi. Berlin: Mouton de Gruyter.Vitale, A. J. 1981. Swahili Syntax. Dordrecht: Foris.Wunderlich, D. 1996. “Minimalist Morphology: The Role of Paradigms”. In G. Booij and

J. van Marle (eds.), Yearbook of Morphology 1995. Dordrecht: Kluwer, 93–114.Wurm, S. A. 1975. “The Trans-Fly Stock”. In S. A. Wurm (ed.), New Guinea Area Lan-

guages and Language Study: Papua Languages and the New Guinea Linguistic Scene,volume 1 of Pacific Linguistics, Series C, 38. Canberra: Australian National University,323–344.

Young, R. W. and W. Morgan. 1998. The Navajo Language: A Grammar and ColloquialDictionary. Albuquerque: University of New Mexico Press.

University of OsnabruckInstitute of Cognitive ScienceKatharinenstrasse 2449078 Osnabruck, Germanye-mail: [email protected]

46

The Interaction of Morphology and Syntax in Aﬃx Orderhome.uni-leipzig.de/jtrommer/papers/ym2.pdf · The Interaction of Morphology and Syntax in Aﬃx Order Jochen Trommer 1 Introduction

Documents